On Thu, 22 Oct 2026 15:45:13 +0800, Stanley Jhu wrote:
> qos_add_test("hce-reset", "ufs", ufs_test_hce_reset, &mcq_test_opts);
> qos_add_test("tmr-handling", "ufs", ufs_test_tmr_handling,
> &mcq_test_opts);
> + qos_add_test("fault-injection-abort", "ufs",
> + ufs_test_fault_injection_abort, &mcq_test_opts);
All three new tests use mcq_test_opts, and UFS_FAULT_INJECT_TAG_ANY is
used only with ssu-timeout. To see what is covered, I removed each of
the following on top of the series. Each build succeeded, and the UFS
qtests still passed 29/29:
- the legacy half of the abort-success branch in ufs_fault_inject_tmr()
(2/8)
- the READ_10/WRITE_10 match for TAG_ANY in ufs_fault_inject_exec_req()
(1/8)
- the body of ufs_fault_inject_sq_cleanup() (2/8)
Could the fault injection cases also run with io_test_opts, and could
one case cover TAG_ANY with a READ(10) below and above
UFS_FAULT_INJECT_MIN_LBA?
Minor:
- 7/8 and 8/8 drop G_GNUC_UNUSED but keep the old continuation indent
of ufs_send_tmr() and ufs_mcq_send_scsi_cmd_async()
(tests/qtest/ufs-test.c:1903-1905 and 1932 after the series). If each
helper is added in the patch that first uses it, 5/8 does not need
G_GNUC_UNUSED.
- 6/8, tests/qtest/ufs-test.c:2014: the comment says the queue
status/pointers "remain 0" after HCE=1, but only the HCE bit is
checked after it.
Daejun
